Do Corporations Manipulate Earnings to Meet or Beat Analysts Forecasts? Evidence from Pension Plan Assumption Changes

Transcription

1 Do Corporations Manipulate Earnings to Meet or Beat Analysts Forecasts? Evidence from Pension Plan Assumption Changes Yul W. Lee College of Business Administration University of Rhode Island Kingston, RI Phone: (401) Fax: (401) Tong Yu College of Business Administration University of Rhode Island Kingston, RI Phone: (401) Fax: (401) Ting Zhang College of Business Administration University of Rhode Island Kingston, RI Phone: (401) Fax: (401) December 2007

2 Do Corporations Manipulate Earnings to Meet or Beat Analysts Forecasts? Evidence from Pension Plan Assumption Changes Abstract This paper presents evidence that corporations boost earnings through opportunistically increasing the expected rate of return on pension plan assets (ERR) to meet or beat analysts expectations, when otherwise the earnings would have missed analyst consensus forecasts. Such earnings management incentives tend to be stronger when the firm earnings would have slightly missed analysts expectations. In addition, this paper documents that the ERR-increase firms earn favorable stock returns in the short-term, subsequent to changes in the ERR. Further examination, however, shows that the longrun stock returns and operating performance of the ERR-increase firms significantly underperform the control firms. JEL Classification: G14; G32 Keywords: Earnings Management; Analyst Forecast; Defined Benefit Pension Plan 1

3 Do Corporations Manipulate Earnings to Meet or Beat Analysts Forecasts? Evidence from Pension Plan Assumption Changes 1. Introduction Corporate managers have strong incentives to report earnings that meet or beat analysts expectations. One reason documented by previous studies is that stock prices respond favorably to meet-or-beat-analyst firms and managers can benefit from increased stock prices, including exercising stock options or receiving large bonus (e.g., Bartov, Givoly, and Hayn, 2002; Kasznik and McNichols, 2002; Degeorge, Patel, and Zeckhauser, 1999). Other reasons, such as managers compensation contracting, reputation with stakeholders, and career concerns also provide incentives for managers to report higher earnings that meet or exceed analyst forecasts (Graham, Harvey, and Rajgopal, 2005; Healy and Wahlen, 1999). On the other hand, Skinner and Sloan (2002) show that missing analyst earnings forecasts, even by a tiny amount, can cause a big drop in stock prices. Matsunaga and Park (2001) document a significant adverse effect on CEO annual cash bonus when firm earnings fall short of analyst consensus forecasts. Therefore, both the apparent benefits to meet or beat analyst earnings forecasts and potential significant costs of missing analysts expectations provide a rationale for why mangers are well motivated to manage earnings upward to exceed the threshold: analysts earnings expectations (Bhojraj, Hribar, and Picconi, 2005). Recent anecdotal evidence has shown that firms sponsoring defined benefit pension plans can manipulate earnings to meet or beat analyst forecasts through changing pension plan assumption, particularly, the expected rate of return on pension plan assets (ERR). This is because changes in the ERR have an important impact on a firm s operating income, while managers enjoy much discretion in setting this assumption, 2

4 according to current accounting standard. To illustrate, Verizon Communications reported total earnings of $389 million in The company would have reported negative earnings, however, had it not included $1.8 billion in pension income. In fact, Verizon's pension plan did not generate any income in 2001 and instead, it lost $3.1 billion. How could Verizon report $1.8 billion in pension income? Management merely increased the ERR by 25 basis points from 9% in 2000 to 9.25% in 2001, which allowed Verizon to report more expected pension income. Higher expected pension income offsets other pension plan costs, which leads to lower pension expenses (net periodic pension cost) on the income statement. With increases in the ERR, the company reported a higher annual EPS of $3.11, beating analysts consensus forecast of $3.01 by $ Take IBM as another example. According to Fortune (2002), IBM reported $11.5 billion in pretax earnings in 2000, among which its pension plan contributed $1.2 billion, accounting for 10% of total pretax earnings. Further analysis shows that about $200 million in earnings were due to the company s increase in the ERR (from 9.5% to 10%, or an increase of 50 basis points). IBM s actual return on plan assets was 3.06% in IBM maintained the ERR at 10% in 2001 and reported $904 million in pension income, representing 13.2% of its total pretax earnings. IBM s ERR is 1% higher than S&P 500 companies median ERR of 9%. Given its huge size of pension plan asset (about $50 billion), a small amount of increase in the ERR would have substantial effect on IBM s earnings. For instance, a 10 basis points increase in the ERR would increase IBM s expected pension income by $500 million. Thanks largely to its increased expected returns on pension plan assets, IBM reported higher earnings and its EPS beat analysts 1 This example is taken from Pay Without Performance: the Unfulfilled Promise of Executive Compensation, BeBchuk and Fried (2004), p Analysts EPS forecast data for Verizon is taken form the I/B/E/S file (both mean and median earnings forecasts were $3.01 in 2001 for Verizon). 3

5 expectations both in 2000 and These two examples clearly illustrate how manager can opportunistically increase the ERR to inflate earnings to meet or beat analyst earnings forecasts. Due to the significant impact of pension plan assumptions on firms financial condition and earnings, the SEC recently has initiated a probe of some big firms, including Delphi, Ford, Navistar International, and Northwest Airlines. According to The Wall Street Journal on November 9, 2005, the SEC asks: did the companies tweak key financial assumptions of these (pension) plans to make the companies themselves look more flush? 2 Motivated by the above anecdotal evidence, we empirically test whether corporations manage the earnings by opportunistically increasing pension plan assumption in this study. In particular, we focus on two research questions: (1) Do managers increase the ERR to meet or exceed analysts expectations when the earnings would otherwise have missed analysts consensus forecasts? (2) How does the market respond to the ERR-increase firms, including the short-term stock returns and the longrun operating performance and stock returns? A number of unique characteristics of defined benefit pension plan have made it an interesting test ground for studying managers earnings manipulation incentive. First, managers enjoy much discretion in setting the ERR. According to current accounting standard (SFAS 87), pension plan asset return is an expected return rather than a realized return. Even when firms actually have negative rate of return on their pension plan assets, managers can still use a higher expected rate of return on the financial statements, as further described in Section 2.2 and 2 Another important pension assumption is pension obligation discount rate, which is used to estimate the present value of pension liabilities. Under the SFAS 87, firms are required to determine the discount rate based on prevailing interest rate for 30-year U.S. Treasury Bill. The latter SEC s rulings and other legislation have largely reduced managers discretion in setting pension discount rate. The focus of this paper is on the ERR, given that the discount rate features considerably less latitude in managers determination. 4

6 Appendix A. Second, pension plan assets and liabilities are large and pension returns have significant effects on a firm s operating performance and earnings. Earnings are sensitive to even a small change in the ERR, particularly for firms with pension plan assets that are large relative to operating income. 3 More importantly, as shown in previous IBM and Verizon cases, increasing the ERR appears to be an effective way to boost firm earnings. By multiplying this expected rate of return by the fair value of pension plan assets at the beginning of the year, firms report expected returns on their plan assets, which offset other components of pension expenses. The higher the ERR, the higher dollar value of the expected returns on plan assets. Ceteris Paribus, the less pension plan expenses (the net periodic pension cost, or NPPC) reported on the income statement, the higher the earnings or EPS (Appendix A). Therefore, managers are able to inflate earnings in the short-term through increasing the ERR. To investigate managers earnings manipulation incentive, we construct a new variable: pseudo-eps. The pseudo-eps is calculated by removing the effect of increases in the ERR from the I/B/E/S actual annual EPS. Because managers are motivated to report higher earnings to meet or exceed analysts forecasts, we would expect such incentive to be stronger if the pseudo-eps is below analyst consensus forecasts. That is, we hypothesize that increases in the ERR are more likely when earnings would otherwise have missed analysts expectations (pseudo-eps < analysts expectations). Consistent with this hypothesis, after controlling for other exogenous factors that determine the changes in the ERR, we find that managers are more likely to increase the ERR when 3 Bergstresser, Desai, and Rauh (BDR, 2006) provide an excellent example to show how a firm s earnings are sensitive to changes in the ERR. Suppose a firm has $100 in operating assets and $20 in pension assets. The firm earns a 4 percent or $4 returns on its operating assets. If the firm changes its expected returns on plan assets from 10 percent to 11 percent, it will increase its net income by $0.20, or 5 percent. This simple example suggests that firms with pension plan assets that are large relative to operating income are much sensitive to the change in the ERR (p7). 5

7 such changes are critical in helping firms boost the pseudo-eps to meet or exceed analysts expectations. Further, we show that such managerial incentive tends to be stronger when the earnings would have slightly missed analyst forecasts, relative to largely missed analyst forecasts. This is because if a firm s earnings are well below analysts forecast, a simple increase in the ERR probably would not help much. In fact, in our sample we estimate that increases in the ERR have helped firms inflate annual EPS by an average (median) of $0.032 ($0.011). Prior studies have documented that a disproportionately large number of firms meet or beat analyst earnings forecasts merely by one or two cents. A common interpretation is that corporations manipulate earnings to meet or beat analysts expectations (e.g., Hayn, 1995; Burgstahler and Dichev, 1997; Degeorge et al., 1999; Burgstahler and Eames, 2003). Our estimated magnitude of earnings management supports this interpretation. Next, we examine how the stock market reacts to those firms that meet or beat analysts expectations through pension plan assumption changes. This is an interesting question because if investors could see through a firm s beating analysts forecasts as a result of earnings manipulation, they would have adjusted inflated earnings and discounted stock prices accordingly. This will substantially reduce managers earnings manipulation incentive. On the other hand, if investors fail to detect firm earnings management the market will react favorably to the ERR-increase firms that meet or beat analysts expectations. In this case, managers are able to capitalize on investors favorable reactions to manipulate earnings. Our findings suggest that investors do not take into account managers earnings manipulation incentive. We find positive stock returns in the short-term following increases in the ERR. For example, the [-1, +3] cumulative abnormal returns adjusted by Fama-French 3-factor-and-momentum model 6

8 (market-adjusted model) are 0.28% (0.27%) for the ERR-increase firms. Further, we regress stock returns on firm earnings announcement effect (proxied by earnings surprises) and an interaction variable between earnings surprises and the ERR changes. We find a significant and positive relation between earnings surprises and the short-term stock returns. Consistent with prior studies on firm beating analysts expectations (e.g., Bartov et al., 2002; Kasznik and McNichols, 2002), this finding of favorable short-term stock returns represents an important benefit for firms managing earnings to exceed analyst forecasts. However, we find an insignificant relation between the interaction variable and stock returns, indicating that investors do not discount stock prices for those meeting-or-beating-analyst firms through changes in the ERR. Our findings also suggest that the positive market reactions to the ERR-increase firms are mainly a magnification of earnings announcement (or earnings surprises) effect. Complementary to the evidence of Teoh, Welch, and Wong (1998a, 1998b) that investors generally do not take into account management incentives to manipulate earnings prior to initial public offerings (IPO) and seasoned equity offering (SEO), our results show that investors do not consider managers opportunistic increase in the pension plan assumption to report higher earnings. Our findings also collaborate with Chan, Chan, Jegadeesh, and Lakonishok (2006) in that the market fixates on bottom-line earnings and neglects other relevant information, which in this case is managers opportunistic increase in the pension plan assumption to inflate earnings in order to meet or beat analysts expectations. In addition, our findings of the short-term market reactions contribute to the literature on the valuation of defined benefit pension plan. Picconi (2006) reports that investors fail to efficiently incorporate pension information that is publicly available into firm valuation. Franzoni and Marin (2006) find that the most underfunded firms earn 7

9 lower stock returns, before and after risk adjustment, in subsequent 5 years than those having overfunded pension plans. Our evidence shows that investors do not differentiate a firm s expected pension earnings from its operating earnings, suggesting that pension earnings are often overvalued by the market (Coronado and Sharpe, 2003). The final test investigates stock returns and operating performance for the ERRincrease firms in the long run. A number of studies suggest that stock price reactions to various events in the short-run are often incomplete. 4 As an important corporate event, the ERR changes offer an interesting setting for the examination of the long-run performance. Our previous results show that ERR-increase firms earn positive stock returns in the short-term, indicating that investors fail to detect firm earnings management through increases in the ERR. The examination of the long-run performance of the ERR-increase firms therefore shall provide insight on market reactions and investors behavior. Using both buy-and-hold abnormal returns and calendar-time portfolio methods to measure the long-run abnormal stock returns, we find that stock returns for the ERR-increase firms significantly underperform our control groups. For instance, the average five-year buy-and-hold abnormal returns for the ERR-increase firms are % and %, relative to size-and-industry-matched ERR-unchanged samples and size-and-bm-matched ERR-unchanged samples, respectively. Further investigation shows that the stock underperformance of the ERR-increase firms is a result of these firms poor operating performance. We find that ERR-increase firms tend to 4 For example, Loughran and Ritter (1995) find that equity issuers substantially underperform for three to five years following the initial offering. Ikenberry, Lakonishok, and Vermaelen (1995) document abnormally high returns after share repurchases, and Ikenberry, Rankine, and Stice (1996) find positive abnormal returns following stock splits. See Daniel, Hirshleifer, and Subrahmanyam (1998) for a summary of this evidence along with a behavioral explanation for the market s initially underreaction to corporate disclosure. 8

10 have poor operating performance and low growth prospect in each of the five years following increases in the ERR, when compared with our control samples. Taken together, this study provides evidence that mangers boost earnings through opportunistically increasing the expected rate of return on pension plan assets to meet or beat analyst earnings forecasts, when otherwise the earnings would have missed analyst consensus forecasts. Such earnings manipulation incentive tends to be stronger when firms earnings would have slightly missed analysts expectations, relative to large earnings missing amount. The ERR-increase firms enjoy favorable stock returns in the short-term, which offers important evidence for the explanation of managerial incentives to inflate earnings through changes in the pension plan assumption. However, the ERRincrease firms experience low stock returns and poor operating performance in the longrun, relative to those control firms that have reduced or maintained the ERR. This suggests that the potential benefit of increasing pension plan assumption has only a temporary positive effect on stock prices. This paper closely relates to a recent study by Bergstresser, Desai, and Rauh (2006, BDR). BDR study investigates a similar line of hypothesis about firms earnings management and pension assumption changes. Our study is different from BDR study in the following aspects. First, we investigate a different motivation for manager s earnings manipulation to meet or beat analyst consensus forecasts; whereas BDR study examines managers incentive to report higher earnings relative to the previous year or higher earnings growth relative to industry peers. Burgstahler and Dichev (1997) and Degeorge et al. (1999) identify three thresholds that managers attempt to exceed: reporting positive earnings, reporting earnings higher than previous year, and meeting or beating analyst earnings forecasts. The focus of this paper is on the third threshold. Prior studies, 9

11 including the BDR study, have extensively investigated the first two thresholds, (e.g., Hayn, 1995; Barth, Elliott, and Finn, 1999). Beginning in the mid-1990s, particularly after Arthur Levitt, the former Chairman of the SEC, delivered his famous remarks The Numbers Game, 5 a growing body of research has shifted to examine the association between earnings management and meeting or beating analysts expectations. Our study therefore complements the BDR evidence and contributes to this stream of research by providing new evidence on managers earning manipulation incentive to meet or beat analysts expectations within a defined benefit pension plan setting. 6 Another difference from BDR study is that we examine market reactions subsequent to changes in the ERR, both in the short-term and long-term, as well as the firm long-run operating performance. As the first paper that has examined the long-run performance of the ERR-increase firms, this study shed further light on investors reactions to earnings management. Our findings of favorable short-term stock returns and long-run stock underperformance for the ERR-increase firms are instructive in better understanding market reactions and investors behaviors. In particular, investors initially fail to detect managers earnings manipulation and the market cheers when these firms report higher earnings meeting or beating analysts expectation. Therefore, we observe favorable stock returns in the short-run. But increases in the ERR can only have a shortterm positive effect on firm earnings. As operating performance deteriorates and its negative impact on a firm s future earnings manifests, investors lower their expectations 5 Increasingly, I have become concerned that the motivation to meet Wall Street earnings expectations may be overriding common sense business practices. Arthur Levitt, former Chairman of the SEC, delivered his remarks The Numbers Game at the NYU Center for Law and Business on September 28, In detecting the frequency and magnitude of earnings management, Chen et al. (2005) find that, among managers incentives to exceed the above three earnings thresholds, the incentive to meet or beat analysts forecasted earnings is the greatest. 10

12 and discount firms stock prices, leading to the stock underformance in the long-run, compared with the control samples. A majority of previous studies on earnings management typically use either aggregate accounting accruals (the difference between earnings and net cash flows from operating activities) or specific accruals as a proxy for the level of earnings management. 7 In contrast, our study directly investigates whether firms employ one of the important pension assumptions as a channel to manage earnings. There are some potential advantages associated with the examination of the pension assumption as an earnings management channel. Accrual-based models largely rely on the measurement and assumption of accruals, making it difficult to capture earnings management through cash flows such as different treatments of R&D expenses (Healy and Wahlen, 1999). The ERR employed in this study is explicit and publicly available on the footnotes of a firm s annual report. It is also readily comparable across firms and free of potential computational issues related to accounting accruals (BDR, 2006). More importantly, the sensitivity of a firm s earnings to the ERR changes provides a unique setting to test the incentive of managerial choices of pension assumption. Our study directly links earnings management with pension assumption changes, an area where an explicit pattern can be identified and therefore, it overcomes the above weaknesses inherent with the accrual based earnings management research. The remainder of the paper is organized as follows. Section 2 discusses defined benefit pension plan background and develops our main hypotheses. Section 3 describes 7 For earnings management research on aggregate accruals, see, e.g., Louis (2004), and Cheng and Warfield (2006). For specific accruals studies, see, e.g., Comprixm Mills, and Schmidt (2006) for interim expenses related accruals (i.e., cost of goods sold, SG&A, and income tax), Liu, Ryan, and Wahlen (1997) for loan loss reserve for the bank, and Beaver and McNichols (1998) for claim loss reserves for insurers. For a comprehensive review on earnings management literature before 1999, see Healy and Wahlen (1999). 11

13 the data and sample. Section 4 reports empirical results on managerial earnings manipulation incentive through increasing the ERR. Section 5 examines the long-run stock returns and operating performance for the ERR-increase firms. We conclude in Section Pension Plan Background and Hypothesis Development 2.1 An Overview of DB Pension Plans 8 In a defined benefit pension plan, employees are entitled at retirement to receive a certain amount of benefit based on their years of service and salaries. Accounting for DB pension plans requires managers to make a number of assumptions, including the expected rate of return on pension plan assets (ERR), the discount rate to be used in estimating projected pension benefit obligations, and employee salary increase rate, etc. The present value of total amount of employees benefits represents a firm s projected pension benefit obligations (PBO). Firms are required by Employee Retirement Income Security Act (ERISA) of 1974 to set aside a certain amount of funds to meet their pension obligations. These funds usually invest in stock and bond markets, and should be managed solely for the interest of employees, not for the interest of employers. The market value of these funds is called the fair value of pension assets (FVPA). If FVPA are larger than PBO, a pension plan is overfunded. If FVPA is lower than PBO, a pension plan is underfunded and firms have to make contributions to their pension plans. This will reduce earnings and adversely affects cash flows. Pension plan assets and obligations are disclosed on the notes to the balance sheet while pension plan expenses are reported on the income statement. 8 This section and Appendix A provide a brief discussion of pension accounting rules. See Dyckman, Davis, and Dukes (2001), Hawkins (2001), and Zion and Carcache (2002) for detailed discussions. 12

14 It is interesting to first look at historical evolution of DB pension plan funding. Panel A of Figure 1 plots the proportion of firms with underfunded pension plans (PBO > FVPA) from 1993 to 2005, including 21,792 firm-year observations. Contemporaneous to the rising market in the late 1990s, about 31% firms were underfunded in 2000, relative to more than 60% underfunded in Pension plans funding status deteriorated significantly in early 2000s when the U.S. stock market declined, with more than 80% firms underfunded in Panel B illustrates the aggregate pension funding surplus (deficit) versus amount recognized on the financial statement. It shows that the aggregate fund surplus increased dramatically since 1996 and peaked in 2000 to $259 billion, concurrent with the booming stock market in the late 1990s. However, a combination of falling stock market and declining interest rate have devastated firms pension plans since 2000 and left firms with a $523 billion in pension deficit in Even after the market rallied later, the aggregate pension fund deficit was still as high as $336 billion in Surprisingly, relative to a substantial amount of pension deficit since 2002, firms have reported gradually increased pension surplus on their balance sheet. For instance, there were totally $66 billion in pension surplus reported on the balance sheet in 2003, much lower than $523 billion in pension deficit. The difference between actual pension surplus/deficit and the amount recognized on the balance sheet can be better understood in terms of economic and accounting meaning of pension funding. The actual pension surplus/deficit represents an economic definition, namely, the difference between market 9 Falling stock market and decreasing interest rate would negatively affect a firm s pension plan funding. On the one hand, falling stock market has caused a sharp decline of FVPA because pension assets are predominantly invested in stocks. According to Pension & Investment Survey (2004), the assets allocation for a typical large DB plan is 63% in equity, 30% in bond, and 7% other. On the other hand, to make things even worse, the decreased interest rate has resulted in a substantial increase of firm s PBO because firms use interest rate to discount their future pension obligations to get PBO. The combined effect is to significantly widen the difference between PBO and FVPA and hence, the funding status of DB plans would deteriorate dramatically. 13

15 value of plan assets and projected benefit obligations. It represents a firm s actual pension funding situation. The amount recognized on the balance sheet simply represents an accounting meaning, which can be significantly different from the actual pension surplus/deficit Earnings Management through Changes in the ERR The Financial Accounting Standards Board (FASB) passed SFAS 87 in 1985 to standardize the accounting rules for defined benefit pension plans. SFAS 87, subsequently amended in 1998 by SFAS 132, requires firms (1) disclose major pension assumptions, including the ERR, the discount rate, and employee compensation increase rate, and (2) report annual pension expenses on the income statement. Lower pension expenses would reduce a firm s total expenses, thus increasing earnings. As shown in Appendix A, pension expenses (net periodic pension cost, or NPPC) consist of four parts: a service cost, an interest cost, other costs, and expected returns on pension plan assets. The expected returns on plan assets, equal to the ERR times the market value of plan assets, offset the interest costs, service costs, and other costs. The higher the ERR, the higher dollar value of the expected returns on plan assets; Ceteris Paribus, the less the net periodic pension cost, or the less pension expenses reported on the income statement. Lower expenses lead to higher earnings or EPS. Furthermore, managers have much discretion in determining the ERR, which could depart significantly from the realized return. Statement of Financial Accounting Standard (SFAS 87): Employers Accounting for Pensions states that the ERR shall reflect the average rate of earnings expected on 10 The complexity of pension accounting and the insufficient disclosure of pension plan information have been subject to criticism for a long time. In September 2006, the FASB released Statement 158, Employers Accounting for Defined Benefit Pension and Other Postretirement Plans. The Statement requires that firms recognize the funded status of defined benefit pension and other postretirement benefit plans in their statement of financial position. SFAS 158 is the outcome of Phase I of FASB s long-term project to reconsider all aspects of pension and other postretirement benefit accounting. 14

16 the funds invested or to be invested to provide for the benefits included in the projected benefit obligation. In estimating that rate, appropriate consideration should be given to the returns being earned by the plan assets in the fund and the rates of return expected to be available for reinvestment (Paragraph 45). In other words, if managers assume the pension plan can earn 10% rate of returns, but in reality the pension plan lost the value, say earn a negative 5% instead, the assumed 10% gain is still used to compute the annual pension expenses; while the difference between the expected and realized returns is amortized over a long period. This briefly explains why firms can boost earnings by merely increasing the ERR. 2.3 Hypothesis Mangers incentive to inflate earnings through increasing the ERR would be heightened if managers anticipate that such changes could help boost earnings to meet or beat analyst consensus forecasts, when the earnings would otherwise have missed the threshold. To empirically investigate the association between the determinants of changes in the ERR and mangers earnings manipulation incentive, we construct a new variable: pseudo-eps, calculated by eliminating the effect of changes in the ERR from the I/B/E/S actual annual EPS. The difference between pseudo-eps and analyst median forecasts, scaled by the absolute value of median forecasts, is pseudo earnings missing amount, or MissAmt. A negative MissAmt (i.e., pseudo-eps is below analyst earnings forecasts) indicates that a firm s earnings would have missed analysts expectations without changing the ERR. Using pseudomiss as a dummy variable to indicate whether a firm s EPS would otherwise have fell short of analyst earnings forecasts (pseudomiss = 1 if MissAmt <0; and 0 if MissAmt 0), we would expect that mangers are more likely to increase the ERR if pseudomiss = 1 than if pseudomiss = 0. If managers are absent from 15

17 earnings management incentives, we would not expect any association between changes in the ERR and the pseudomiss. Further, we expect that managers earnings management incentives tend to be stronger when the earnings would have been slightly below analysts consensus forecasts, when compared to large earnings missing amount. If pseudo-eps is substantially below analyst earnings expectations (i.e., MissAmt is more negative), a simply increase in the ERR would not help firms boost earnings much. The above analysis leads to our main hypotheses as follows: H1: Ceteris Paribus, managers have strong incentive to increase the expected rate of return on pension plan assets when they anticipate the actual EPS would otherwise have missed analysts expectations. H2: Ceteris Paribus, managers earnings management incentive tends to be stronger when the earnings would have slightly missed analyst earnings forecasts, relative to largely missed analyst earnings forecasts. 3. Data and Sample The data comes primarily from three sources. The financial statement data about DB pension plan is from Compustat from 1993 to Data on stock returns is from CRSP file. Analysts earnings forecasts data is obtained from the I/B/E/S database. 3.1 Pension Plan Data and Variables The initial pension sample consists of all firms available on Compustat file from 1993 to 2005 that sponsor a DB pension plan. There are three important pension variables of interest: the fair value of plan assets (FVPA), the present value of pension obligations (PBO), and the ERR. Following Franzoni and Marin (2006) and Picconi (2006), we set FVPA as the sum of overfunded pension plan assets (item 287) and underfunded pension 16

18 plan assets (item 296), and set PBO as the sum of overfunded pension obligations (item 286) and underfunded pension obligations (item 294). Consistent with Amir and Benartzi (1998) and the BDR study, the ERR is set to equal to pension benefit-anticipated longterm rate of return on pension assets (item 336). We classify a firm as a sponsor of DB pension plans if it has FVPA and PBO information available in Compustat file. The ERR data is available in Compustat from 1991 forward. We choose 1993 as our beginning year to ensure that all firms in the sample have the ERR (item 336) available in Compustat. We exclude financial institutions (SIC between 6000 and 6999) and utilities (SIC between 4400 and 5000) because managers in these regulated industries might have different motivations to manage earnings (Burgstahler and Eames, 2003). After identifying firms with DB pension plans, we define a firm as pension overfunded (underfunded) if its FVPA is larger (smaller) than the PBO. In other words, a firm has plan surplus (deficit) if the difference between FVPA and PBO is positive (negative). Pension funding ratio is then calculated by scaling plan surplus (deficit) with firm total market value at the beginning of fiscal year. 11 We follow Franzoni and Marin (2006) selection criteria to screen the sample. First, a firm must have at least 2- year pension accounting data available in Compustat. This requirement helps correct for the survival bias induced by the way Compustat constructs data (Banz and Breen, 1986). Second, stock price at the fiscal year end must be higher than $1. Third, we winsorize pension funding ratios at the top and bottom 1% to control for the influence of outliers. Our pension sample covers from 1993 to 2005, and includes 21,792 firm-year 11 Other scalars in calculating pension funding ratio include total assets, operating income, employee number, and fair value of pension assets (FVPA). Following Franzoni and Marin (2006), we normalize pension surplus/deficit by market value in this study. We use other measures for robustness check. 17

19 observations, with the average of 1,671 firms in each year. The year with the maximum (minimum) number of firms is 2004 (1993) with 1,803 (1,552) firms. 3.2 Analyst Earnings Forecasts Data Analyst earnings forecasts data is from the I/B/E/S unadjusted summary statistics file. Analysts issue multiple forecasts for a firm and we use the most recent forecast proceeding to the annual earning report date because previous studies (e.g., O Brien, 1988) have found that the recent forecasts are more accurate. The median forecasts are used as analysts consensus forecasts because medians are less sensitive to outliers. Given the precision in the decimal places required to calculate MissAmt, we use the unadjusted I/B/E/S summary forecasts data to avoid losing the precision of measurement due to the I/B/E/S adjustments of prior forecasts for subsequent stock splits (Baber and Kang, 2002; Payne and Thomas, 2003). Firms followed by fewer than two analysts are deleted (Mendenhall, 2003). After combining pension data with analyst earnings forecast data, our final sample is reduced to 12,513 firm-years, among which there are 1,503 ERR-increase firm-years, 3,212 ERR-decrease firm-years, and 7,798 ERR-unchanged firm-years. 3.3 Descriptive Statistics We partition our sample into three groups based on the direction of changes in the ERR: (1) increase group; (2) decrease group; and (3) maintain or unchanged group, with the latter two groups served as our control groups, matched based on industry, size, and book-to-market value. A firm is classified as an ERR-increase firm if its ERR in year t is greater than that in previous year t-1, and is classified as an ERR-decrease (unchanged) firm if its ERR in year t is less than (equal to) that in previous year t-1. We report our 18

20 findings for the ERR-increase group along with those for two control groups. Wherever possible, we attempt to draw comparisons between our findings for these three groups. The evidence from control groups will further help us evaluate empirical results for the ERR-increase group. Table 1 reports descriptive statistics and the distribution of the ERR. Panel A presents the percent of firms with the ERR in each specified range for each year from 1993 to The table shows that ERR varies significantly across firm-years, ranging from below 6% to above 13%, but a majority of firms have ERR between 8% and 10%. About 34.7% (7.0%) of firms in 1993 reports the ERR between 8% and 9% (7% to 8%) while this percentage increases to 56.9% (17.5%) in About 19% of firms in 1993 reported the ERR higher than 10% while this percentage decreased to less than 1% in Overall, about 76.1% of all firms have the ERR between 8-10%. Panel B reports the summary statistics of the ERR in the sample. The mean (median) of ERR is 8.60% (8.86%), with the mode of 9.00% across the sample period. The mean (median) of ERR decreases from 8.93% (9.00%) in 1993 to 7.92% (8.25%) in Particularly, the mean (median) ERR maintains at around 8.7% (9%) from 1993 to 2001, and it has started to decline since The ERR reaches the lowest level in The mode for the ERR in both 2004 and 2005 is 8.5%, down from 9% in previous years. The summary statistics show that on average, firms have gradually decreased the ERR. However, consistent with BDR study, we find that some firms do increase the ERR through the sample period, as further shown in the following section. Panel A of Table 2 reports pension plan characteristics of firms changing the ERR. It shows that the ERR-increase group generally has larger pension assets and pension obligations than the ERR-unchanged group. For instance, the average pension 19

21 plan asset (FVPA) for the ERR-increase firms is $1.05 billion, almost double that of ERR-unchanged firms ($0.61 billion). The ERR-increase and decrease groups are close in the size of pension assets and obligations. Notably, the ERR-increase firms have worse (more negative) pension funding ratios. This group also recognizes more pension assets than the ERR-decrease group ($38.78 million vs. $24.65 million) on the financial statement. Panel B reports top 10 industries for the ERR-increase group based on Fama- French industry classification code. The most frequent industry for the ERR-increase is telecommunication, with 84 firms or 4.51% of all the firms in the sample, followed by retail (2.58%) and machinery industries (2.52%) Empirical Results on Earnings Management through the ERR Increase 4.1 Frequency of Changes in the ERR Table 3 reports how frequently the sample firms with DB pension plans change the ERR from 1993 to The percentage of firms that increase, decrease or maintain the ERR is presented relative to (1) the previous year, and (2) the year of Panel A shows that from 1993 to 2005, there are totally 1,864 (3,814) firms that increase (decrease) the ERR for at least once, or 8.58% (17.55%) of total observations, with a majority maintaining the ERR. Similar findings are also reported in BDR study. The average increase (decrease) amount of ERR is 71.6 (76.6) basis points. With the average pension plan assets of $1.05 billion (Panel A of Table 2), such an increase in the ERR can be translated into about $7.2 million in pension income increase on average. Panel A also shows that increases in the ERR are more common in the late 1990s and early 2000s. More than 10% firms in each year from 1996 to 2001 have increased the ERR. Given that 12 We control for the industry effect in the following regression analysis. 20

22 the late 1990s represents the market booming period and that pension plans usually have a large portion of funds investing into the stock market, the high frequency of the ERR increases during this period appears to be justifiable. During the period in which stock market declined significantly, however, there are still a relatively large portion of firms increasing the ERR. For instance, 13.48% (11.82%) of firms in 2001 (2000) increased the ERR, with an average of 64.7 (80) basis point increase. Panel B shows that over a longer period of time, about 10.96% (64.10%) of firms have increased (decreased) the ERR relative to In relative to the year of 1992, 2,623 firms increase their ERR, representing 10.96% of total firms in the sample. For the following empirical tests, the ERR-increase firms refer to those firms that change the ERR relative to the previous year, rather than relative to the year of Univariate Analysis of the ERR Changes Table 4 presents our preliminary evidence showing that increases in the ERR can help firms report higher EPS. The average actual FCE, or the forecasted earnings errors measured as the I/B/E/S actual annual EPS minus the most recent analysts median forecasts preceding the announcement date, scaled by the absolute value of median forecasts, are less negative than those for the control groups in each year from 1993 to A couple of years (1995, 2000, and ) actually have positive FCE for the ERR-increase group. The FCE on average is -2.04% for the ERR-increase group, relative to -10.8% for the ERR-decrease firms, and -9.87% for the ERR-unchanged firms. Although we cannot attribute less negative FCE completely to firms increasing the ERR at this point, a further examination of pseudo-fce for the ERR-increase group reveals some interesting findings. After taking out the effect of increasing the ERR on a firm s actual EPS, the pseudo-eps, as expected, has become much less than the actual EPS, 21

23 resulting in larger forecast earnings errors. The average pseudo-fce, calculated as the difference between the pseudo-eps and the median earnings forecasts, scaled by the absolute value of analysts median forecasts, are -5.03%, which is more negative than the actual FCE of -2.04%. This result provides the preliminary evidence that increases in the ERR can help firms boost the EPS (i.e., reduce the forecast errors). 4.3 Regression Analysis Managers can increase, decrease or maintain the pension plan assumption. We use a logistic regression model to test the association between changes in the ERR and managers earnings manipulation incentive. In particular, we are interested in whether mangers increase the ERR to boost earnings when their earnings would otherwise have missed analysts earnings forecasts. We construct a new variable: pseudo-eps, which is calculated by eliminating the effect of changes in the ERR from the I/B/E/S actual annual EPS. 13 With the pseudo-eps, we are able to calculate MissAmt, or the difference between pseudo-eps and analysts median forecasted EPS. Specifically, ERR FVPA 0.65 pseudo EPS I / B / E / S Actual Annual EPS (1) Common SharesUsed to Calculate EPS pseudo EPS I / B / E / S Analyst Median Forecasted EPS MissAmt (2) AbsoluteValue of Analyst Median Forecasted EPS We use a dummy variable, PseudoMiss to indicate whether the pseudo-eps misses or beats analysts expectations: PseudoMiss = 1 if MissAmt < 0 (pseudo-eps < analysts median forecasted EPS) PseudoMiss = 0 if MissAmt 0 (pseudo-eps analysts median forecasted EPS) 13 We assume that (1) other components of pension costs, including interest cost, service costs are unchanged; and (2) the effective tax rate is 35% for all firms in the sample. 22

24 Our first hypothesis states that managers are more likely to increase the ERR when they anticipate the earnings would have been below analyst forecasts. The following regression equation is designed: where: ΔERR i,t = α + β 1 (PseudoMiss i,t )+ β 2 (Sensitivity i,t )+ β 3 (MissSensitivity i,t ) + β 4 (FRatio i,t )+ β 5 (Acq_SEODummy i,t )+ β 6 ( ARRDummy i,t-1 ) + β 7 (LeadERR i,t-1 ) + ε i,t (3) ΔERR i,t =ERR i,t ERR t-1 PseudoMiss i,t Sensitivity i,t MissSensitivity i,t FRatio i,t Acq_SEODummy i,t ARRDummy i,t-1 LeadERR i,t-1 =1 if pseudo-eps is misses analysts forecasts; and 0 if it meets or beats analysts forecasts = log (Pension Assets i,t /Operating Income after deprecation i,t) =an interaction variable of PseudoMiss and Sensitivity = pension funding ratio, (FVPA i,t PBO i,t)/market value i,t = 1 if firm i conducts any acquisition activities or issues seasoned equity offerings (SEO) in year t, and 0 if otherwise = 1 if firm i s actual rate of returns on pension plan assets in year t-1 increased relative to actual rate of returns on pension plan assets in year t-2; and 0 otherwise = firm i s expected rate of returns on pension plan assets in previous year t-1 The dependent variable ΔERR is defined as follows: ΔERR = 1 if management increases the ERR in year t relative to year t-1; ΔERR = 0 if management maintains or decreases the ERR in year t relative to previous year t-1; Alternatively, the ΔERR takes the value of 1, 0, and : 14 We measure the dependent variable ERR as the annual change of the ERR between the year t and t-1. The annual changes include both unexpected and expected changes in the ERR. We would prefer our dependent variable to measure only unexpected changes in the pension assumptions so that we can attribute those changes to managers earnings management. However, we cannot directly observe the unexpected 23

25 ΔERR = 1 if management increases the ERR in year t relative to year t-1; ΔERR = 0 if management maintains the ERR in year relative to year t-1; ΔERR = -1 if management decreases the ERR in year relative to year t-1. The dummy variable PseudoMiss i,t captures potential managerial motivation to inflate earnings to meet or exceed analysts expectations. Our hypothesis states that mangers are more likely to increase the ERR if pseudo-eps is below analysts consensus earnings forecasts (PseudoMiss =1). We thus predict that β 1 > 0 in equation (3). Sensitivity i,t is included into the model to measure operating income sensitivity to changes in the ERR. As reported in BDR study, management has significant discretion in setting the ERR and their earnings management incentive would become stronger when the operating income is sensitive to the changes in the ERR (i.e., a small increase in the ERR resulting in a big increase in operating income). 15 In addition, we include an interaction variable between PseudoMiss i,t and Sensitivity i,t into the equation. The interaction variable identifies the effect when the ERR changes would have a larger impact on the firm earnings. We include a number of control variables in the equation based on prior studies: (1) FRatio: pension funding ratio. A firm with a high (low) funding ratio will be less (more) likely to change the ERR for the earnings management purpose; (2) Acq_SEODummy: a dummy variable indicating whether a firm conducts acquisitions or issues seasoned equity offerings (SEO). BDR study reports that firms are more likely to increase the ERR when managers prepare to acquire other firms or to issue SEO; (3) ARRDummy: an indicator of whether a firm s actual rate of return on pension plan changes. Following the BDR study, we use the difference between current ERR and prior year s ERR as the measure of our dependent variable. 15 See note 3. 24

26 assets in year t-1 increases relative to actual return in year t-2. If a firm s pension plan assets earn a higher realized rate of returns (ARR) in previous year, managers are more likely to increase the expected rate of returns (ERR) for current year; and (4) LeadERR: the ERR in the prior year t-1. A firm s choices of the ERR might be affected by the prior year s ERR. This is because a firm with a high (low) ERR in the prior year will have less (more) room to further increase the ERR than a firm with a low (high) ERR. The second hypothesis states that managers earnings manipulation incentive tends to be stronger when earnings would have marginally missed analyst earnings forecast, relative to large earnings missing amount. We slightly modify the equation (3) as follows to test this hypothesis: where: ΔERR i,t = α + β 1 MissAmt i,t + β 2 (Sensitivity i,t )+ β 3 (MAmtSensitivity i,t ) + β 4 (FRatio i,t )+ β 5 (Acq_SEODummy i,t )+ β 6 ( ARRDummy i,t-1 ) + β 7 (LeadERR i,t-1 ) + ε i,t (4) MissAmt i,t MAmtSensitivity i,t = the different between pseudo-eps and analyst forecasted EPS, scaled by absolute value of analysts forecasts. See equation (2). = an interaction variable between MissAmt and Sensitivity Other variables are defined the same as before. Different from the test for equation (3), the sample used to run the equation (4) only includes those firms with PseudoMiss = 1 (MissAmt < 0); or those firms that would have missed analysts earnings forecasts after eliminating the effect of the ERR increase on the actual EPS. We do not include those firms with pseudomiss = 0 (MissAmt 0) because managers tend to have less or no incentive to increase the ERR to report higher EPS, given that the PseudoEPS already meets or beats analysts expectations without changing the ERR. We are particularly interested in these PseudoEPS -missing (MissAmt < 0) firms because 25